kNN classifiers 1.2. Improving resource efficiency

Instance-based learners such as the kNN classifier implemented in Praat are fairly inefficient as far as CPU and memory usage is concerned. The performance can be improved upon by making sure that only those instances that are vital for the accuracy of the classifier are stored and that non-vital instances are disposed of. Praat does give the user the possibility to prune non-vital or harmful instances, making the resulting classifier less memory and CPU hungry and in some cases more accurate even though that is not the primary objective of the pruning algorithm.

Links to this page


© Ola Söder, May 29, 2008